A user has built "Acorn," a self-contained 3U inference server housing multiple GPUs for distinct AI workloads. The system employs two RTX 6000 Blackwell MaxQ GPUs for a mid-size MoE LLM (GLM-5.3-flash), an RTX 2000E for embedding and reranking, and an RTX 4000 SFF for voice ASR and TTS, backed by 256GB of ECC UDIMM RAM. After six months of development, the author reports the private server is operational and generating practical value beyond a prototype stage.